Hiring an AI Consultant? Ask This First

Ask me how to choose an AI consultant and I won't start with credentials or client logos. I start with one question: what would this person tell you not to build? The consultants worth hiring spend the first call finding reasons to say no; the ones who aren't spend it finding reasons to say yes — and that one tell matters more than anything on their website.

Why there are suddenly so many AI consultants

The label got cheap this year. Agentic AI and the Model Context Protocol went from research-lab jargon to shipping infrastructure — MCP alone has crossed ten thousand public servers, and most major platforms now ship it by default. Building a demo that looks impressive takes a weekend and an API key. That combination — cheap infrastructure, cheap demos, real hype — is exactly what turns a specialism into a costume. Marketing freelancers doing prompt tweaks call themselves AI consultants. IT resellers bolt "AI-powered" onto an existing pitch deck. One-person shops running someone else's playbook set up a website in an afternoon.

None of that is dishonest, exactly. But it means the label "AI consultant" now carries almost no signal by itself. A few years ago, the filter question was "do they even understand this technology." That filter is gone — plenty of people understand the technology. The harder, more useful question now is who understands your business well enough to talk you out of a bad idea, and that is a completely different skill from knowing how a transformer works.

I don't think this is a bad thing for the market overall — cheaper tooling and a lower barrier to entry mean more genuinely useful small shops exist now than three years ago, and plenty of the newcomers are good. But it does mean the buyer has to do more of the filtering work themselves, because the market can no longer do it for them. Nobody is going to hand you a certified list of the honest ones. The best you can do is ask better questions than "have you built AI before," because at this point almost everyone will say yes.

What an AI consultant actually does

Ask five people what an AI consultant does and you'll get five different answers, which is itself part of the problem. In practice, the job sits at an intersection of three things: translating a business problem into a decision a model can actually support, choosing between building something custom, buying an off-the-shelf tool, or integrating what you already have, and then staying around long enough to see whether it changes anything.

That's a genuinely different job from the categories it gets lumped in with. A chatbot or automation vendor sells you their product and is, quite reasonably, motivated to make your problem look like a fit for it. An IT integrator wires systems together competently but rarely questions whether the underlying process should exist at all — that's not what they're hired to do. A freelance machine-learning engineer is often excellent at the technical core but usually has no mandate, and no real interest, in touching your process or your org chart. An AI consultant's actual job is to sit across all three and be willing to say the AI part isn't the hard part — the hard part is almost always the decision, the data and the people around it.

The three questions worth asking in a first call

If you take one thing from this essay, take these three questions into your next vendor call.

• "What would you tell me not to build?" If the honest answer is "nothing, everything you mentioned sounds great," that is the single clearest red flag available to you. Every real project has at least one idea that should be shelved, delayed or radically shrunk.
• "What happens to this after you leave?" Ask who owns the code, the prompts and the data pipeline, and whether your own team could run it without them. If the answer depends on their proprietary platform, you haven't bought a solution — you've rented a dependency.
• "Which of your last five projects got killed after the pilot, and why?" A consultant who can't name one either hasn't been doing this long enough to have a failure, or isn't being straight with you. I ask myself the mirrored version of all three before I ever put a proposal in front of a client, because a client who can't get honest answers to these deserves better than us too.

AI consultant, IT company, or freelance engineer?

The three options solve different problems, and picking the wrong one is a bigger risk than picking the wrong AI model. A freelance ML engineer is the right call when you already know precisely what to build and just need skilled hands to build it — you've done the thinking, you need the execution. An IT company or systems integrator is the right call when the work is mostly plumbing: connecting tools, migrating data, keeping a thin layer of AI on top of infrastructure that already works. An AI consultant earns their fee when you're genuinely unsure which of several ideas is worth doing, or when the project touches process, people and technology all at once — which, in most SMEs, is most of the time. We've written up a longer, more mechanical comparison of boutique consultancies against the big-four firms and a similar breakdown of hiring an agency versus a freelance ML engineer, if you want the fuller decision tree.

Does your company even need one yet

Size changes the answer more than most people expect. A one- or two-person business is usually better served by an off-the-shelf tool — a good chatbot, a bookkeeping integration, a workflow app — than by any consultant, because the fee alone can exceed what the project is worth; the right conversation there often ends with "don't hire anyone, buy the tool." A small business in the twenty-to-fifty-person range is typically the sweet spot: big enough that a real process (quotes, invoicing, customer service, scheduling) is costing real hours every week, small enough that one well-scoped project can be felt company-wide within a quarter. A mid-sized organisation moving from a pilot into production has a different problem again — less "should we do this" and more "how do we govern it, who owns it internally, and how do we manage the change" once more than one team depends on it.

Match the engagement to the size of the problem, not the size of the hype. I've sat across the table from micro-businesses convinced they needed a bespoke agentic system when a €30-a-month tool would have solved it, and from fifty-person firms trying to run a serious automation programme through a single freelancer with no capacity to support it once it broke. Both mismatches are avoidable with one honest sentence early in the conversation: "here's what a project your size actually needs, and here's what it doesn't."

The 2026 twist: your shortlist forms before anyone calls

Here's something genuinely new I've started noticing in first conversations this year. Prospects increasingly arrive with a shortlist that's already half-formed — not from a Google search, but from a conversation they had with ChatGPT, Claude or Gemini about who does this kind of work well. The evaluation used to start the moment someone opened a search engine. Now it starts earlier, inside a chat window, often before a human at the company has consciously decided to start looking.

That's a real shift in how trust gets built, and it rewards a different kind of honesty than a polished homepage does. What an AI assistant tends to surface and quote back is plain-language explanation of what a firm actually does, honest write-ups of where a certain approach doesn't work, and a track record specific enough to be quotable rather than vague enough to be forgettable. That's not a coincidence — it's the same reason we publish an honest, named comparison of AI consultants in the Netherlands rather than a generic "why choose us" page: the buyers doing this kind of research before their first call, including with us, are exactly the ones worth having that first call with.

The pitch itself is the biggest tell

You can usually tell more from how a consultant sells than from what they sell. Watch for a quote or fixed roadmap delivered before they've seen a line of your actual data. Watch for a demo that only works cleanly on their example and starts stumbling the moment you push back with your own numbers. Watch for nobody on the call who can describe, specifically, a project that didn't work and why — not as a gotcha, but because a firm that has never had a project fail either hasn't done enough of them or isn't telling you the truth. And watch for anyone who answers "what does this cost" with a number before they've asked you a single question back — a real scoping conversation runs both directions, and a price that arrives before the questions do is usually a guess wearing a quote's clothing. I've written before about how I can usually tell within the first half hour whether a project is going to work, and the signals are nearly identical to the ones that tell you whether the person across the table is worth trusting with it.

The inverse pattern is just as telling. The best conversations I have start with a client describing a problem and me spending most of the call asking questions rather than answering them — and sometimes the honest answer at the end of that call is that they shouldn't build anything yet, which is a harder thing for a consultant to say out loud than it sounds, and sometimes genuinely the right one.

What happens after the contract is signed

The evaluation shouldn't stop at go-live. Ask what maintenance costs once the project is running, who is on the hook when the underlying model gets an upgrade or a price change, and whether you are locked into one vendor's API or genuinely able to swap the model behind your own layer without a rebuild. A consultant worth keeping around treats the model itself as the least important, most replaceable part of the whole engagement — everything else is what has to survive when a better model inevitably ships.

A quick word on price

Cost deserves its own honest treatment rather than a paragraph buried in a hiring essay — we've laid out realistic ranges and what actually drives them on our AI project cost page — but the one thing worth saying here is that price alone tells you almost nothing about quality. I've seen quotes for the same problem differ by a factor of five, and the expensive one was sometimes the worse fit. What should move the number is scope, ownership and how much of the thinking has already been done before the proposal lands in your inbox, not the size of the logo on the website.

Where to start

If you're not yet sure whether you need an agency, a freelancer, or nothing at all this quarter, start with a real business question rather than a vendor call — the order matters more than the choice of provider. When you are ready to have that first conversation, look at how a firm talks about its own limits before you look at its client list; that single habit tells you more than a portfolio ever will. You can see how we approach that first conversation ourselves when you're ready to have it.

Frequently asked questions